Round Corner
Department of Computer and Information Science


BIO-INSPIRED: Understanding and Visualizing Deep Convolutional Networks

SINTEF Ocean is leading a research project – Intelligent- aiming to develop novel concepts for salmon behaviour prediction in net cages. The concepts and methods developed in Intelligent inspired by state-of-the-art BigData concepts and exploits Deep Learning to enable improved context awareness and intelligence for monitoring and control of operations in aquaculture industry. An interesting area of research is Computer Vision and the use of Convolutional Neural Networks(CNNs). These networks learn to represent images through learning increasingly advanced filters as the network grows deeper. Recently, video classification tasks have also seen improvements through the use of CNNs and this has also been explored in the Intelligent project.
In this project assignment the student will investigate what Convolutional Neural Networks are actually learning through training on video data. The student will have access to a large dataset of underwater videos of salmon in net cages. The student will analyse and investigate the filters in several trained deep CNNs to investigate what the networks are learning from the videos and to vizualize the outputs of each layer in the networks. One example deep learning architecture the student will analyze in detail is the 3D-CNN architecture. This architecture is able to capture temporal information from videos by using stacks of video frames as inputs. This enables the network to learn more useful features from videos.
The assignment is as following:
- Using the dataset of existing videos and using existing trained models, explore and visualize what the networks have learned and how this impacts the model performance.
- Train new models, using the results from the analysis, to improve model performance on prediction accuracy.
A state-of-the art workstation with the necessary software to work with videos will be made available to the student.
- Excellent programming skills in Python
- Knowledge and interest in deep learning, machine learning and computer vision concepts.
Co-supervisors from SINTEF Ocean:
Dr. Ekrem Misimi – Senior Scientist ( and Håkon Måløy

The work may continue to Master Thesis

  IMPORTANT: If you sign up for this project, please send a) your CV (including a transcript with all of your college grades, and b) a brief explanation of WHY you want to do this particular project to Prof. Keith Downing (


Keith Downing Keith Downing
308 IT-bygget
735 90271 
NTNU logo